[Breaking News] Google “Gemini 3.8 Flash” released.

Small Flash overtook large Frontier ・HLE-Verified 54.9% exceeds GPT-5.6 Sol(54.5%)/Claude Opus 5(54.4%) ・Input $0.75/output $3.75 (1M tok・until 12/31) ・1M context ・64k output Video = Official demo of creating a DOS version of Google Maps with 1 prompt

TORASep 2

Gemini 3.8 Flash at the top, matching Opus 5 with a 74% score on DeepSWE v1.1

Md Ismail Šojal 🕷️Sep 2

Gemini 3.8 Flash now ranks ahead of Opus 5 on DeepSWE-bench.

It shows particular strength in front-end coding and agentic tasks, while having a fast response time. You can run it via the `agy` harness inside Orca, where it works well paired with complementary models like Fable

Orca ADESep 252

This quietly ends the whole claude vs gpt argument

Moon dev: "openai's newest model is literally the best at math on the planet and i can prove it in one screen" "i'm a die hard claude guy and even i can't ignore what just happened to opus 5"

stacksbyraySep 2